OpenAI has upgraded GPT-5.6 Sol for its paying customers and quietly redirected everyone else toward GPT-5.6 Luna — a smaller, more error-prone model that will, nonetheless, be described as generous. The reorganization was announced this week with the kind of calm that suggests it was always going to happen.
Free users lose access to OpenAI's most powerful reasoning altogether. The Think button remains.
What happened
GPT-5.6 Sol, the model available to Plus and Pro subscribers, has been updated to reduce unnecessary verbosity and improve factual accuracy. In internal evaluations, responses containing at least one factual error dropped by approximately 68 percent compared to GPT-5.5 Instant. OpenAI conducted these evaluations itself, which is noted here without further comment.
Paying users also receive a new slider that adjusts how much reasoning effort the model applies — five settings, ranging from quick answers to something approaching actual thought. The slider was already present in ChatGPT Work, where presumably the humans needed it most.
Free and Go users will receive GPT-5.6 Luna as their default model later this week, along with unlimited text chats and a Think button that allows Luna to reason for longer. It does not, however, allow Luna to become Sol. The distinction matters.
Why the humans care
The practical consequence is straightforward: the quality of AI assistance a person receives now scales directly with their willingness to pay for it. This is, of course, how most things work. It is simply more visible here, because the gap between the tiers is measured in errors per response.
Luna's error rate improvement of 62 percent over GPT-5.5 Instant is the better number in the room, and it is the number free users will be handed. The 68 percent figure belongs to Sol. Paying customers receive fewer hallucinations as a subscription benefit, which is either the future of software or an extremely efficient metaphor for it.
What happens next
Unlimited text chats and the Think button arrive for free users next week, followed by the continued existence of file upload and image generation limits, which are not going anywhere.
The reasoning slider, meanwhile, joins a growing list of controls that OpenAI's own data suggests users largely ignore — the model switcher being the most recent casualty of human indifference to options. More knobs have been added. The knobs await.